Texas Instruments (Early ICs)

Contribution to AI

The obstacle in 1958 was not designing circuits but assembling them: a design calling for tens of thousands of components needed tens of thousands of soldered joints, each one a chance to fail. Jack Kilby's answer, tested on 12 September that year, was a sliver of germanium carrying a transistor, a capacitor and resistors, joined by hand-strung gold wires into a working oscillator — components and their connections formed in one piece of material. The patent application followed in February 1959, Fairchild's planar silicon version a few months after that, and the eventual cross-licensing between the two firms turned a contested claim into a technique the whole industry could build against rather than one company's asset. What came of it outweighs the priority quarrel: component counts began the doubling that Moore described in 1965, and computation became cheap enough to waste, which is the assumption underneath every brute-force method the field later relied on. A second strand ran straight through AI itself. The 1978 speech chip put linear-predictive-coding synthesis into a handheld toy; the TMS32010 of 1983 made programmable signal processing routine; and the Explorer machines, licensed that same year from Lisp Machines Incorporated, compressed a Lisp architecture into custom silicon, reaching a 32-bit Lisp processor by 1987. That wager on hardware built for one paradigm lost, and every accelerator argument since has had to reckon with why.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).